Microsoft adds token usage to model cards

Raw performance benchmarks are becoming secondary to economic viability; the developer community is moving toward cost-to-performance efficiency as the primary metric for deploying AI in production.

• –**Economic Pragmatism:** In production, developers want the cheapest model that clears their quality threshold rather than the most expensive frontier model.
• –**Exposing Inefficient Architectures:** Standardizing "average token usage" will highlight which models achieve high scores through inefficient multi-step prompting or agentic looping versus those with optimized native reasoning.
• –**Pressuring Competitors:** Microsoft's move will likely compel other major cloud and model providers to disclose similar token consumption metrics to remain competitive.